The simplest elements we can observe are very specific parts of the body and how natural they look.

• Eye movement: especially in videos, unblinking eyes are very revealing. Algorithms learn  from photos, and in photos found online, people don’t often close their eyes. Also, eye movements tend to follow the speaker, so it’s difficult to reproduce them realistically.
• Facial expressions: These can appear unnatural and may not express emotion at all.
• Body movements: if they’re jerky, if certain parts don’t move in sync with others, this can help you spot a fake video. The authors of less elaborate deepfakes focus mainly on the face, so it’s not uncommon for body movements or positions to be unnatural enough to make it easy to spot.
• Shadows: The way light hits the face can easily reveal a fake image or video, as shadows will fall in a way that’s inconsistent with the scenery/position.
• Hair that’s too perfect: deepfakes tend to produce images with perfect hairstyles; the algorithm doesn’t necessarily recognize, and therefore doesn’t learn unruly hair.
• Unnatural teeth: currently, the algorithms don’t seem to be able to generate slightly imperfect teeth, as this would require going tooth by tooth; deepfakes therefore have a “denture effect”.
• Inconsistent sounds: Deepfakers generally pay more attention to the image or sound.
• A video in slow motion can sometimes be used to detect desynchronization between speech and lip movements.
• Looking at the image on a larger screen: Deepfakes are often designed for people watching from their phones. On a larger screen, such as a computer monitor, details may be easier to see.

It is also possible to use reverse image search to find similar images online; for films, there is currently no publicly available reverse video search tool.